Anthropic Blocks Chinese Users from Circumventing Access Restrictions to Claude
0xBroomberg
Anthropic is systematically shutting down the workarounds Chinese companies use to access Claude — from cloud-service relays to offshore subsidiaries — with Ant Group and ByteDance both named, marking another hard wall rising between frontier AI tools and China.
How did Chinese companies access Claude in the first place?
Ant Group gave employees Claude enterprise access through its Singapore subsidiary's corporate intranet.
ByteDance took a different route: it launched a reimbursement program this year, letting engineers subscribe to Claude personally, expense the cost, and connect via VPN.
Neither method breaks any U.S. or Chinese law — but both violate Anthropic's terms of service. This means → Anthropic is not playing a legal card but a contract card: my product, my rules on who gets to use it.
What role does Microsoft Azure play?
Another pathway: Chinese firms accessed Claude's API through Microsoft Azure cloud services, using offshore subsidiaries as the contracting entity.
Microsoft said "Anthropic is responsible for monitoring usage and enforcing its terms; Microsoft provides support." In plain terms = Microsoft kicked the ball back to Anthropic and positioned itself as just the pipe.
A person familiar with the matter called offshore-entity access "a known issue" that is not limited to any single cloud provider.
Are there other grey-market channels?
A category of "relay" services exists: they route requests from mainland Chinese users through overseas-registered Claude accounts and return the results.
But major Chinese AI companies typically avoid these relays — not for compliance reasons, but for commercial security. They worry operators may store or resell prompts, letting competitors reverse-engineer their usage patterns.
This reflects a telling reality: for big companies, the data-leak risk outweighs the ban risk.
Why are Chinese engineers so determined to use Claude?
The core driver is distillation — using a more capable model's output to train a smaller model, so the smaller one mimics the larger one's abilities.
Anthropic's coding tools are especially popular among Chinese software engineers and AI startups; Claude's output quality is a critical input for distillation results.
This means → Chinese engineers are not chasing a chatbot — they are chasing a low-cost shortcut to boosting their own models' capabilities.
What makes Anthropic claim it is the strictest?
Anthropic says it is the only frontier AI company that explicitly restricts sales to Chinese-controlled entities — including subsidiaries registered outside China.
Its detection methods include an "evolving detection system" that has used Claude Code to check the time zone on a user's computer, among other signals, to determine if the user is physically in mainland China.
In plain terms = it does not just check your registration address — it checks your computer's clock. Technical enforcement has drilled down to the device level.
Where does this cat-and-mouse game end?
On the Chinese regulatory side: Beijing has banned domestic firms from using offshore models to build consumer-facing applications, but allows AI labs to use foreign models for internal R&D.
This means → China's own rules leave a door open for "internal R&D use" — and Anthropic's rules try to shut even that door.
As Anthropic keeps upgrading its detection, whether Chinese engineers can maintain access to top-tier U.S. AI tools will remain a running variable in the U.S.–China AI capability race.
Content is for reference only, not financial advice.